An Efficient Neighborhood-based Interaction Model for Recommendation on Heterogeneous Graph

Abstract

There is an influx of heterogeneous information network (HIN) based recommender systems in recent years since HIN is capable of characterizing complex graphs and contains rich semantics. Although the existing approaches have achieved performance improvement, while practical, they still face the following problems. On one hand, most existing HIN-based methods rely on explicit path reachability to leverage path-based semantic relatedness between users and items, e.g., metapath-based similarities. These methods are hard to use and integrate since path connections are sparse or noisy, and are often of different lengths. On the other hand, other graph-based methods aim to learn effective heterogeneous network representations by compressing node together with its neighborhood information into single embedding before prediction. This weakly coupled manner in modeling overlooks the rich interactions among nodes, which introduces an early summarization issue. In this paper, we propose an end-to-end Neighborhood-based Interaction Model for Recommendation (NIRec) to address above problems. Specifically, we first analyze the significance of learning interactions in HINs and then propose a novel formulation to capture the interactive patterns between each pair of nodes through their metapath-guided neighborhoods. Then, to explore complex interactions between metapaths and deal with the learning complexity on large-scale networks, we formulate interaction in a convolutional way and learn efficiently with fast Fourier transform. The extensive experiments on four different types of heterogeneous graphs demonstrate the performance gains of NIRec comparing with state-of-the-arts. To the best of our knowledge, this is the first work providing an efficient neighborhood-based interaction model in the HIN-based recommendations. CCS CONCEPTS • Information systems → Data mining; • Computer systems organization → Heterogeneous (hybrid) systems.

Learning Intents behind Interactions with Knowledge Graph for Recommendation

“…Clearly, the recommendation accuracy heavily relies on the quality of paths. However, two mainstream path extraction methods suffer from some inherent limitations: (1) Applying brute-force search easily leads to laborintensive and time-consuming feature engineering, when largescale graphs are involved [44]; (2) When using meta-path patterns to filter path instances, it requires domain experts to predefine the domain-specific patterns, thus resulting in poor transferability to different domains [15,17].…”

“…However, the sparse reward signals, huge action spaces, and policy gradientbased optimization make these networks hard to train and converge to a stable and satisfying solution [50,52]. [17,38,39,41,47] are founded upon the information aggregation mechanism of graph neural networks (GNNs) [13,14,19,34,42]. Typically, it incorporates information from the one-hop nodes to update the representations of ego nodes; when recursively performing such propagations, information from multi-hop nodes can be encoded in the representations.…”

“…CKAN [47] uses two different strategies to separately spread collaborative signals and knowledge-aware signals. More recently, NIRec [17] is proposed to combine path-and GNN-based models, which propagates interactive patterns between two nodes through meta-path-guided neighborhoods.…”

“…Some follow-on studies [15,44,49] enrich the interactions with multi-hop paths from users to items for better characterizing user-item relations. However, these methods struggle to obtain high-quality paths, suffering from various issues like labor-intensive feature engineering [44], poor transferability to different domains [15,17], and/or unstable performance [49]. More recently, a technical trend [38,39,41,47] is to develop end-to-end models founded on graph neural networks (GNNs) [9,13,19,34].…”

Initialization Matters

“…From the aspect of recommendation techniques, existing methods are mainly based on collaborative filtering [38,40,47] and deep learning [17,22,59]. Graph Neural Networks (GNN) have recently gained popularity in recommendation systems [5,37,56].…”